RAM specifications with AM2

[Thornblade]

I saw the following review about an AM2 dual core chip at new egg:

"Alot of people don't know the truth about Socket AM2. For Socket AM2 to gain huge, you have to use expensive low-timed RAM. If you want to see 15%-20% speed increases over Socket 939, you will need to buy 3-3-3-8 DDR2-800. This may cost alot for most people, but the performance is worth it for a step up to K8L and 4x4 or 8x4 in the future."

Can someone please explain to me what 3-3-3-8 is? And where would I look for this number when I'm buying RAM? It doesn't say in the review what this attribute of the RAM is called, just the numbers.

Thanks.

[Fozzik]

Well... first I think it's important to note that the person who wrote that review is smoking something, and they aren't sharing. :confused:

From everything I've seen and read, what that person is saying is not correct.

Second, I've discussed what RAM timings are all about in a lot of detail in some old threads, I'll try to dig that up for you (search is your friend!). Not all manufacturers of RAM will list the timings, and you can probably guess that if they don't, the timings probably aren't very good (most of them do list them now).

Ah...here's a post that should help -
http://www.silkyvenom.com/forums/showpost.php?p=3593&postcount=11

[Thornblade]

Thanks for the link, Fozz. I'm trying to figure out - based on what you wrote in that post - the smaller the numbers are then the better? Simple-minded explanations work best for a simpleton like myself. :confused: I guess I'm trying to figure out how to tell which RAM chip works better than another.

May I also ask - what is wrong about what that person wrote? I mean, I don't know enough to determine that (obviously), but I think it was written in a response to a lot of people reviewing the chip and saying it wasn't much faster than their 939 socket chips.

And, also, based on your post, I gather then that the attribute you would call these "numbers", is the DRAM address??

(search is your friend

Thanks for the suggestion. Since I had no idea what to call these numbers, I had no idea what to search for.

Thank you again for your help.

[akherat]

There was a recent article over at toms hardware (http://www.tomshardware.com/2006/05/23/amd_reinvents_itself/)about this very subject -

The long and short of it is this: you need to use a 2.6Ghz or higher AM2 processor with CAS4 or lower memory to see any bang for your buck (today) with DDR2.

In general terms the lower the CAS latency of the RAM, the "faster" it is, and as we all know, faster is better :)

Even though you may not fully understand it, the article I linked over at Toms is a really interesting look into the AM2 - and if you like doing research, there are a number of other articles over there discussing AMD, RAM and chipsets and whatnot that are quite enlightening.

[rabb1t]

Thanks for the link, Fozz. I'm trying to figure out - based on what you wrote in that post - the smaller the numbers are then the better? Simple-minded explanations work best for a simpleton like myself.

Think of those numbers as timing a trip from place a to place b. So yes, the smaller numbers are better, as it represents a faster round trip. The overall ram rating (ex: DDR2-800) is like how fast something can go overall, so in this case the bigger numbers are better.

As to your original question, sure 3-3-3-8 timed ram would kick the butt of 5-5-5-12 timed ram. However, in terms of real world performance you'd see maybe an 8-10% gain in terms of framerates. (And that is likely overly generous, it is probably more like 5-7% gain.) So let's assume you have a game running at 60 FPS on the slower ram, that would be 66 FPS on the 'tighter' ram. Is that really worth it?

Well, you'd say, 'sure that sounds worth it', until you hear the price. Currently, this Corsair XMS2 2 gig (http://www.newegg.com/Product/Product.asp?Item=N82E16820145035) kit at 3-4-3-9 is the fastest DDR2-800 I've seen. It's price, $469. Compare that to the average timed Corsair XMS2 2 gig (http://www.newegg.com/Product/Product.asp?Item=N82E16820145590) kit at 5-5-5-12 timing for $177 after rebate (up until this week it was $158 after rebate) and you can quickly see what I mean. (I have never seen the high-end ram go on sale.)

Recap...
$469 for 3-4-3-9 timing
$177 for 5-5-5-12 timing
Assuming a ~10% gain in speed, this costs you ~260% in price.

Is probably what would be less than a 10% speed gain really worth more than 2.5x the cost? Well, that is up to you to decide, but I can guarantee if you are looking for framerates you would likely see a far greater gain by spending that $ on a higher end GPU. (As example a 7900 GT is ~$300 and the 7950 GX2 is $550. For that ~175% cost you'd be getting a 50-100% increase. Far above a 0-10% gain from the ram.)

Sure, I’d say upgrade your planned ram timing, but only as a last step after the GPU power and CPU speeds, and only if you have unlimited spending. Otherwise the gain simply isn't worth the cost.

[Fozzik]

The problem I had with the quote is that anytime someone says that RAM timings (or any one thing inside the computer, for that matter) is going to give you HUGE gains in performance...they are either ill-informed or smoking something.

Overall performance in a computer system is a combination of the performance of all the parts, and the software being run. It's very complex, with tons of factors constantly changing the equation...and it is extremely rare that a single factor (like RAM timings, which is a smaller than average factor anyway) would change overall performance more than slightly.

Looks like rabb1t covered things pretty well. Although...that's a bit of a stretch on the discription of RAM clock speed. :)

Clock speed is the metronome inside the computer. It's the drum beat that all the devices march to. So, RAM which is clocked at 800 MHz (effectively) is hearing a drum beat 800,000,000 times a second. Latency is measured in the number of individual clock cycles it takes to actually accomplish things.

So, if it takes you 4 clock cycles to strobe between columns in the RAM... you are talking about clock speed divided by latency to get the response time. This is still an over-simplification...but I think it's easier to understand.

Example -
The memory is DDR2-800, and has a CAS latency of 4.

clock speed / latency = how many times a second.
800,000,000 / 4 = 200,000,000

So the RAM can switch between columns 200 million times a second. This is all semi-abstract, and because there are a LOT of different timings that come into play, the differences that latency makes are always going to be smaller than it seems like they should. For instance, change the CAS latency in our example above to 3 instead of 4. This would increase the number of column changes to 267 million instead of 200. Big difference...which doesn't end up translating very well into real world performance.